Repository navigation
[Doc] [History Server] Add doc for running cluster support - #67504
machichima wants to merge 4 commits into
Conversation
Signed-off-by: machichima <nary12321@gmail.com>
Signed-off-by: machichima <nary12321@gmail.com>
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request updates the Ray History Server documentation to explain how it supports accessing and debugging running Ray workload resources in addition to terminated ones. It details periodic log uploads, the new RAY_COLLECTOR_LOG_UPLOAD_INTERVAL environment variable, log chunk storage layouts, and how to access and reload running clusters. The review feedback suggests updating a placeholder in an example URL to <SELECTED_SESSION_ID> for consistency with other examples in the document.
| The History Server caches a session after the first load, so the view doesn't update on its own. To pick up newly uploaded events and logs, click the reload button on the cluster selection page, or add `?reload=true`: | ||
|
|
||
| ```text | ||
| http://localhost:8265/enter_cluster/default/rayjob/rayjob-historyserver-gcs/<SESSION_ID>?reload=true |
There was a problem hiding this comment.
For consistency with the previous example at line 487, consider using <SELECTED_SESSION_ID> instead of <SESSION_ID>.
| http://localhost:8265/enter_cluster/default/rayjob/rayjob-historyserver-gcs/<SESSION_ID>?reload=true | |
| http://localhost:8265/enter_cluster/default/rayjob/rayjob-historyserver-gcs/<SELECTED_SESSION_ID>?reload=true |
dstrodtman
left a comment
dstrodtman
left a comment
There was a problem hiding this comment.
Docs-side style review from Douglas Strodtman (Anyscale docs team). Claude Code assisted with this review; I read and stand behind every comment. Scope is style, grammar, and Ray docs conventions only (writing-style guide). The KubeRay maintainers own whether the described behavior is right.
This is a clean update. The new section reads well, and the claims I spot-checked match kuberay-project/kuberay#5347 (the 5m default and the <file>.chunks/<offset> layout). A few small things inline. Two of them are lines outside the diff that still say the History Server is for terminated workloads only. I folded those into the nearest in-diff comment.
I agree with holding the merge as the description says. The reload button, ?reload=true on the selection page, and the automatic reload after deletion come from kuberay-project/kuberay#5463 and #5467, which are both still open.
No objection from the docs side once the inline items are addressed.
Signed-off-by: machichima <nary12321@gmail.com>
|
Thanks @dstrodtman for the review! I fixed all your comments |
Description
kuberay-project/kuberay#5347 added running cluster support by uploading log chunks to the object store. Reload button and auto reload after RayCluster terminate are added in kuberay-project/kuberay#5463 and kuberay-project/kuberay#5467 respectively.
This PR update the history server document for running cluster.
NOTE: we should merge this after kuberay-project/kuberay#5463 and kuberay-project/kuberay#5467 are merged.
Related issues
Part of kuberay-project/kuberay#5452
Additional information